Image denoising using multivariate model in shiftable complex directional pyramid domain and principal neighborhood dictionary in spatial domain

The major challenge for image denoising is how to effectively remove the noise and preserve the detail information to get better visual quality and higher peak signal-to-noise ratio (PSNR). A new image denoising methods based on combination of multivariate shrinkage model in shiftable complex directional pyramid (PDTDFB) domain and principal neighborhood dictionary (PND) non-local means algorithm in spatial domain is proposed. In PDTDFB domain, the PDTDFB coefficients are modeled as multivariate non-Gaussian distribution taking into account the interscale and intrascale dependency correlation. Then a multivariate shrinkage function is derived by the maximum a posterior (MAP) estimator and the denoised coefficients are obtained. Although the PDTDFB-based algorithm achieves efficient denoising result, it is prone to producing salient artifacts which relate to the structure of the PDTDFB. Principal neighborhood dictionary (PND) is further employed to alleviate the artifacts with small computational load in spatial domain. Experimental results indicate that the proposed method is competitive with other excellent denoising methods in terms of PSNR value and visual quality.